Quantifying Effort in American Football

Emily Shteynberg, Luke Snavely, Sheryl Solorzano

Background and Motivation

  • Not all rushing yards are equal

  • Traditional stats miss the “how” behind yards gained

  • Previous research has explored athletes reaching theoretical max capacity 1

  • Can we “measure” effort using tracking data?

  • Multiple ways of evaluating effort

    • Intangible and subjective
    • Depends on player position, defense faced, game context, snap count/play volume, play call/assignment

Our Data: 2022 NFL Season

  • Game, play, player, tracking data from Weeks 1-9 1

  • Running plays where a running back (RB) is the ball carrier

  • Trimmed each play to frames between handoff and end of play

Motivation: acceleration-speed (AS) profiles tell us about players’ acceleration capacities at different speeds (Morin et al., 2021)

Effort = percentage of a player’s points (frames) above the relaxed regression line

  • Biased toward backups because of lower sample size…or are they just working harder? Are starters fatigued or pacing themselves?

  • Unrealistic theoretical max speeds

Quantile regression provides a better way of evaluating players’ acceleration capacities

New ways of looking at effort

Effort v1

\[ \left(\sum\limits_{i=1}^{n_{\text{below}}} {\frac{1}{1+d_i}}\right)\bigg/n_{\text{below}} \] ➜️ Quantifies how close a player comes to his “best” (99th percentile) accelerations

➜ Saquon Barkley: 0.152

➜ Rex Burkhead: 0.149

Effort v2

Percentage of total points that lie in between the percentile \(P_{99}\) and \(P_{99}-3\)

➜ Quantifies how often a player comes close to his “best” (99th percentile) accelerations

➜ Saquon Barkley: 0.074

➜ Rex Burkhead: 0.069

Concerns and limitations

  • Unsure about threshold for relaxed percentile line in Effort v2

  • Problem with qgam: some players have more than 1% of points above 99th percentile

    • Account for this without penalizing players who simply have large raw number of points above the line

Plan of action

  • Define research question and scope ✅

  • Data cleaning and preprocessing, EDA ✅

  • Develop basic AS profiles for players ✅

  • Obtain effort metric(s) with quantile regression ⏳

  • Evaluate effort metric(s) by correlating it to effort-related outcomes